I work as a systems developer at a large food manufacturing firm in Australia. Recently we have been fighting with Microsoft's Analysis Services about a certain cube we were wanting processed.
The quick details -
Database size: 25.6Gb
Dimensions: 9
Storage Model: MOLAP
The data was the entire set of sales figures for the years 2002 to present in weekly installments. We had to process this cube during off-peak hours as the server it was running on was a high-demand corporate server. Taking this into account we set it to run at 5:30pm (when most of the people using the server had gone home for the day) and were expecting it to be processed in the morning.
Arriving in the morning we saw a large red cross and a clever error message: "Error: General network error!"
Well - after a few weeks of fiddling and trying to process it we made it to the "Successful" stage - process time: 50 hours.
No way could we run it at that, so we started cutting things down. In the end we removed 3 unnecessary dimensions and culled the data up to 2005. Then switching to ROLAP for better aggregation of older data we doubled the size of the database with the aggregations. That failed miserably, so we took it back to MOLAP.
These are the statistical differences between the two methods:
Old Cube Setup
ROLAP
At 50% performance gain, 235 aggregations @ 3,434.6Mb
At 75% performance gain, 351 aggregations @ 16,509.7Mb
At 100% performance gain, 3624 aggregations @ 258,271.7Mb
MOLAP
At 50% performance gain, 235 aggregations @ 1,412.2Mb
At 75% performance gain, 351 aggregations @ 6,562.4Mb
At 100% performance gain, 3624 aggregations @ 99,826.5Mb
New Cube Setup
ROLAP
At 50% performance gain, 120 aggregations @ 1,052.9Mb
At 75% performance gain, 189 aggregations @ 5,700.9Mb
At 100% performance gain, 2079 aggregations @ 91,681.4Mb
MOLAP
At 50% performance gain, 120 aggregations @ 546.6Mb
At 75% performance gain, 189 aggregations @ 2,921.8Mb
At 100% performance gain, 2079 aggregations @ 45,820.4Mb
Astounding results on a 25 Gb database aren't they?
Well - the intricacies of MOLAP vs. ROLAP are well documented. However here are some empirical results which actually outline the size difference between the two storage models.
For more information on Molap and Rolap data, see ROLAP and MOLAP data
As closure, the new cube processes in about 12 hours and query performance is reduced from 16 minutes to about 1 - 2 mins.